When using a time series training set, do you consider the same object at different times as 2 separate objects?

Let's say I have a stock that I own and the value of that stock over 4 days. I also know if the stock on that day falls into one of 2 categories. On each day I want to make a feature out of the percentage difference in the price of that stock from yesterday.

So for the sake of this example we have 4 data points:

S1 - Jan 1, 2019 - -5% - 1

S1 - Jan 2, 2019 - 4% - 0

S1 - Jan 3, 2019 - 2% - 1

S1 - Jan 4, 2019 - 4% - 0

I want to use a time split to train my model, so I will begin by only training on the the first 2 days, validating on the 3rd, and testing on the 4th. Is it correct to include both the 1st and 2nd as separate inputs in my training set?

So far that is what I have done, and it has given me results that are a little too good to be true

• “On each day I want to make a feature out of the percentage difference in the price of that stock from yesterday.” Can you elaborate on this? – The Laconic Jan 16 at 3:29
• @TheLaconic For example on the 31st if the value was 1 and the 1st it is 2, the value of that feature on the 1st would be 100%. Then on the 2nd if it is 3, the value would be 50%. The point is more that it is a feature dependent on a previous point in the time series – Written Jan 16 at 3:45
• I'm still confused. "Feature" connotes "predictor". But it sounds like this feature is the return on the stock. What are the predictors and what is the outcome you are trying to predict? – The Laconic Jan 16 at 14:52
• @TheLaconic I am attempting to use the percentage difference in the stock from day to day as a predictor for a binary classification into 1 of 2 categories. Note each data point contains a percentage difference and a 1 or 0 denoting if it falls into one of those categories – Written Jan 16 at 15:01